How to Read a Data Center Feasibility Report

Before reading the verdict, find the page that defines the target project. A feasibility study answers whether a site can support a specific load, on a specific schedule, for a specific kind of facility. A site judged feasible for 40 MW of colocation can fail at 300 MW, and a site judged infeasible for energization next year can work with a later date.

• Load: peak MW, the ramp schedule and whether the figure is IT load or total facility demand.
• Facility type, building count and cooling approach, which drive water and land needs.
• Target dates: first power, full build and any deadline in a land contract.
• Scope: which workstreams were covered, which were excluded and what was done in the field versus on the desktop.
• Date of the report and the date of each major data source.

If the target is missing or vague, the verdict is too. Ask the author to restate it before relying on anything else.

Formats vary, but most reports follow a similar structure. Read the parts in a different order than they are written: verdict and open items first, evidence second, narrative last.

Most reports land on one of four verdicts: feasible, feasible with conditions, not feasible, or not yet determinable. The second is the most common and the easiest to misread. “Feasible with conditions” means the site works only if specific things happen, such as a utility building a substation, a rezoning passing or a wetland boundary landing where the desktop data suggests. Each condition is a risk you would be taking on.

The U.S. intelligence community’s analytic standards offer a useful model for reading the rest. ICD 203 requires analysts to express likelihood with a defined set of probability terms tied to numeric ranges, and to state confidence in a judgment separately, based on the quantity and quality of the evidence and the logic behind it.¹ It also asks products to explain what drives uncertainty and to identify indicators that would change it.¹ Site reports rarely use that vocabulary, but the discipline transfers directly.

Watch for words that sound firm but carry no defined meaning: “should,” “appears,” “no known issues,” “ample capacity.” Ask what evidence sits behind each one and how confident the author is in it.

Every finding should say where it came from. A label like “mapped,” “informal utility indication” or “field-verified” tells you more than the finding itself. For example, the National Wetlands Inventory is described by its publisher as reconnaissance-level and not intended for site-specific planning or regulatory use, so a wetlands finding based only on it is low-confidence by design.⁷ Our guide to desktop screening vs. field assessment covers which layers can be trusted for what.

Findings also age. Under the federal All Appropriate Inquiries rule, a Phase I ESA’s site visit, interviews, lien search and records review must be completed within 180 days before acquisition, and the inquiry as a whole must be conducted within one year before acquisition.⁴ An approved jurisdictional determination from the Corps can be relied on for five years, while a preliminary JD is advisory.⁵

Utility and market information ages faster still. A capacity indication from a year ago may not survive a wave of new large-load requests on the same substation, so note the date of every utility conversation.

Power usually decides the verdict, and it is where reports most often overstate certainty. Sort every power statement into one of four levels: mapped (lines and substations near the site), informal indication (a conversation with the utility), study result (a formal load or interconnection study) or agreement (a signed service or construction agreement).

Utilities are tightening what they will say before studies. Silicon Valley Power’s general manager has said the utility used to provide will-serve letters and stopped doing so.⁸ City Utilities of Springfield, Missouri, says every large-load proposal will undergo a system impact study, larger projects also need Southwest Power Pool evaluation, and developers pay the upfront costs of serving them.⁶ Our guide to utility will-serve letters explains what such letters do and do not commit.

Context matters too. Utility pipelines are full of requests that will never be built: Utility Dive reported in 2026 that many utilities and regulators have turned to large-load tariffs with minimum contract terms, collateral, upfront study payments and exit fees to weed out speculative load.⁹ Wood Mackenzie found that utilities in ERCOT were committing to large loads equivalent to 62% of the grid’s 2024 peak demand, and noted that even committed capacity faces regulatory, supply chain and price risk.²

The rules are also moving. On June 18, 2026, FERC issued show cause orders to six RTOs and ISOs, preliminarily finding that their tariffs may not adequately address large and co-located loads.¹⁰ Each operator had 60 days to show its tariff was just and reasonable or propose reforms, including study paths that evaluate generators and the co-located loads they serve together; responses were due in August 2026.¹¹ A report written before those proceedings conclude should say so, and its power timeline should be read as provisional. The large-load interconnection process guide covers the steps.

Feasibility-stage costs are rough by nature. AACE International’s cost estimate classification places Class 5 estimates at 0–2% project definition, used for concept screening and typically built from parametric models, analogy or judgment.³ In AACE’s process-industry practice, Class 5 accuracy ranges run from -20% to -50% on the low side and +30% to +100% on the high side, stated at an 80% confidence interval.³ One public agency’s evaluation notes that Class 4 estimates are prepared on very limited information, have wide accuracy ranges and are typically used for screening and feasibility.¹²

AACE also cautions that the accuracy range should come from a risk analysis of the specific project, not be assumed from the class.³ In practice, a single-point cost figure in a feasibility report is a warning sign. Ask for the estimate class, the range and the three or four items that drive it, which on data center land are usually transmission upgrades, substation work, grading and off-site utilities. See site-driven construction cost drivers for more.

Schedules deserve the same treatment. The critical path is almost always power, then entitlement. A schedule that shows energization before a utility study is complete is a target, not an estimate.

A short list of questions will tell you quickly whether a report can carry a decision.

1. 01What project, load and dates was the verdict written for, and would it change at half or double the load?
2. 02Which findings were field-verified, which were confirmed by the utility or county, and which are desktop-only?
3. 03What is your confidence in each major finding, and what would raise or lower it?
4. 04Who at the utility was contacted, when, and what exactly did they say? Is anything in writing?
5. 05Which conditions behind the verdict are outside our control, and what would each cost or delay if it went the wrong way?
6. 06What is the estimate class and range for any cost figures?
7. 07Which items in the open-items list should be funded first, and why?

If the report leaves these unanswered, the verdict is less certain than it looks. Compare it with how developers frame the same issues in underwriting a site, and use why data center sites fail as a check on what may be missing. For an independent read of a site or a second opinion on a report, you can get a site reviewed.